Research On Algorithm Of Identification And Quantitative Analysis Of Atmospheric Biological Components

Atmospheric biological components refer to small and bioactive particles suspended in the atmosphere,also known as bioaerosols.They are generally composed of aggregations of single cells or microorganisms.Bioaerosols are important parts of atmospheric aerosols.In addition to characteristics of general aerosols affecting climate change,they are also associated with infectivity and sensitization with hiding transmission routes.They are potential threats to environment and health.With the development of the economy,many countries have a demand for long-term,wide-area,real-time monitoring and identification of atmospheric biological components to achieve rapid and accurate warning.Fluorescence lidar combines lidar technology with Laser-induced fluorescence technology,taking into account the high spatial and temporal resolution of lidar.The recognition and analysis ability of Laser-induced fluorescence spectrum is the most likely technology to realize remote real-time monitoring of atmospheric biological components at present.This paper focuses on the algorithms of Fluorescence lidar monitoring of atmospheric biological components,identification and quantitative analysis.The simulated echo spectral signals of three mixed atmospheric biological components verifies the feasibility of the quantitative identification and analysis algorithm for component proportions,and compares the performance of different algorithms.Firstly,the detection principle of fluorescence lidar is briefly described.According to the echo signal,the detection is judged based on whether there are suspected atmospheric biological components in the detected cloud cluster,and further detect the Laser-induced fluorescence(LIF)spectrum of the suspected biological component cloud cluster.Secondly,based on the analysis of echo spectrum signal processing of suspected biological component clouds,mixed fluorescence spectra are extracted from the echo spectrum signal through preprocessing algorithms such as removing background noise,extracting nitrogen Raman signal to correct fluorescence spectra,and filtering processing.We specifically compare the denoising effects of various filtering algorithms such as sliding window smoothing algorithm,wavelet threshold denoising,and empirical mode decomposition,and ultimately chose SavitzkyGolay smoothing algorithm to filter the echo spectral signal.Finally,an improved genetic algorithm combining with vertex component analysis and four supervised learning algorithm models combining with known proportion spectrum are constructed to solve the ratio coefficients of components of the mixed fluorescence spectrum.The algorithm models include principal component regression,back-propagation neural network,one-dimensional convolutional neural network and two-dimensional convolutional neural network.The two-dimensional convolutional neural network model has the best prediction for the ratio coefficients and recognition accuracy,with R2 between the predicted ratio coefficients and the preset ratio coefficients of the three components reaching 0.9919,0.9952,and 0.9917,respectively.At the same time,binary encoding and binary decoding methods are used to identify the spectral composition categories of the mixed spectra based on the obtained ratio coefficients of different components.When the threshold value of binary fitting of two-dimensional Convolutional neural network model combined with genetic algorithm is 2.01%,the recognition accuracy of spectral composition categories can reach 92%.This paper further studies the echo spectral signals under different interference conditions,measuring the generalization ability and anti-interference ability of each algorithm model structure.The results show that the supervised learning algorithm is more anti-interference.Through algorithm design,this article ultimately achieves the monitoring,identification,and quantitative analysis of atmospheric biological components,achieves the goal of providing reliable reference algorithm solutions for fluorescence lidar.